The last piece of Mike Dobson on the complexity of disambiguating places on maps is hard stuff.
Save your time and don't read it if you still want to understand maps as just shiny tiles quickly sent to your screen. 

Otherwise - make sure to also read Mike Blumenthals comment at the original article Dobson is referring, expanding further into the broad range of issues (typos left as quoted):

It is a huge problem, Google gathering data from Streetview cars, governement data, parcel data and user content does a pretty good job when they good data sets.

I have heard that there is a “distance decay” function in map data…the less population & the furhter away from urban areas the worse the data. If Google doesn’t have good streetview data and there is little user generated content, as you have pointed out the government data is suspect then the outcome will not be that great.

 Dobson gets unusually angry in his last paragraphs:

Maybe Google should hire some geographers? Nah, all you need is Algorithms. Well, maybe you need algorithms and somebody who understand how they should be associated. Perhaps that’s no longer in vogue at Google?

That's a true notion certainly. Algorithms and engineers are one part of the equation. Making sense of geography still needs some subject matter expertise beyond algorithmic wizardness.

We had the story earlier in June last year:

Google Mapping Nairobi inhouse ... what's next?

Ed Parsons has a story based on news from NetworkWorld.
Googles' recently opened Nairobi office suspectedly didn't find its own location on Google Maps - and kick-started a new Nairobi city map that didn't exist before. (...)

In June 2008, our open question was: "Will Google follow the OpenStreetMap or is this a one-timer?"

Yesterday, the Mapping News Blog reports (via Slashgeo, kudos!):

"Google are now allowing you to download the vector data (Kenya only so far) in either Keyhole Markup Language (KML) or in ESRI Shapefile format.

Google seems to now return data from Google Map Maker contributors to the public. It's only those datasets that can be released because bindings with Tele Atlas permit for only this. Anything else effectively would kick TA out of business. As the idea behind Map Maker was to engage users to map areas of the World where the duopol is limited or non existent has proven to work regionally and in Nairobi.

Anyway, only non-commercial-non-profit entities seem to be legally entitled to use the data. The fundamental issue of "Who owns the data?" isn't solved - just shifted.

Visit the original site as it has a lot of more information.

update: Mikel Maron at Brain Off has a deeper analysis and insight: "First Thoughts on Google MapMaker Download"